Submodular Benchmark Selection

Alexander Smola

Evaluating large language models across many benchmarks is expensive, yet many benchmarks are highly correlated. We formalize the selection of a small, informative subset as submod…